BioSim—a biomedical character-based problem solving ... WEBSITE/images/publications/BioSim … · ronment for biological problem solving, where it pro-vides a real-time interactive

Future Generation Computer Systems 21 (2005) 1145–1156

BioSim—a biomedical character-based problem solvingenvironment

Yang Caia, Ingo Snelb, Betty Chenga, B. Suman Bharathib, c,Clementine Kleind, Judith Klein-Seetharamana, b, c, e, ∗

a School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USAb Institute for Organic Chemistry, University of Frankfurt, 60439 Frankfurt, Germany

c Institute for Structural Biology, Research Institute Jülich, 52425 Jülich, Germanyd Berufskolleg Kartäuserwall, Abteilung Medien, Kartäuserwall 30, 50678 Köln, Germany

e Department of Pharmacology, University of Pittsburgh School of Medicine, Pittsburgh, PA 15261, USA

Available online 19 May 2004

Abstract

Understanding and solving biomedical problems requires insight into the complex interactions between the components ofbiomedical systems by domain and non-domain experts. This is challenging because of the enormous amount of data andknowledge in this domain. Therefore, non-traditional educational tools have been developed such as a biological storytellingsystem, animations of biomedical processes and concepts, and interactive virtual laboratories. The next-generation problemsolving tools need to be more interactive to include users with any background, while remaining sufficiently flexible to targeto ion of thev nt for theb -playing“ llaborationb io-behaviori recordingso nmentalc -dynamicp tem and ab red bloodc reative andi©

K

0

pen research problems at any level of abstraction, from the conformational changes of a protein to the interactarious biochemical pathways in our body. Here, we present an interactive and visual problem solving environmeiomedical domain. We designed a biological world model, in which users can explore biological interactions by rolecharacters” such as cells and molecules or as an observer in a “shielded vessel”, both with the option of networked coetween simultaneous users. The system architecture of these “characters” contains four main components: (1) b

s modeled using cellular automata; (2) bio-morphing uses vision-based shape tracking techniques to learn fromf real biological dynamics; (3) bio-sensing is based on molecular principles of recognition to identify objects, enviroonditions and progression in a process; (4) bio-dynamics implements mathematical models of cell growth and fluidroperties of biological solutions. The principles are implemented in a simple world model of the human vascular sysiomedical problem that involves an infection by Neisseria meningitides where the biological characters are white andells and Neisseria cells. Our case studies show that the problem solving environment can inspire user’s strategic, cnnovative thinking.

2004 Elsevier B.V. All rights reserved.

eywords:Scientific visualization; Biological discovery; Game design; Problem solving; Artificial life; Education

∗ Corresponding author. Tel.: +1 412 383 7325; fax: +1 412 648 1945.E-mail address:[email protected] (J. Klein-Seetharaman).

167-739X/$ – see front matter © 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.future.2004.04.002

1146 Y. Cai et al. / Future Generation Computer Systems 21 (2005) 1145–1156

1. Introduction

Rapid advances in the convergence of Nanotech-nology, Biotechnology, Information Technology, andCognitive Science (NBIC) are believed to have the po-tential to result in a “comprehensive understanding ofthe structure and behavior of matter from the nanoscaleup to the most complex system”[1]. In biomedicine,high-throughput methodology now allows the accumu-lation of unprecedented amounts of scientific data, suchas genome sequences, gene expression profiles, struc-tural and functional proteomic data. These advanceshave stirred great hopes for understanding and cur-ing diseases, but the quantity of data demands con-vergence with information technology to interpret andutilize these data to advance human performance andquality of life [1]. This requires an understanding ofthe complex interactions between the components ofbiomedical systems by both domain and non-domainexperts. This is particularly challenging in the biomed-ical domain because of the massive data and knowl-edge accumulation. Facilitating convergence of NBICtechnologies therefore requires a novel problem solv-ing environment (PSE) for the biomedical domain.

A biomedical PSE should be a computer systemthat provides all the computational facilities neededto immerse into a biomedical problem. It should hidethe intricacies of computer modeling of physical phe-nomena so that the user can concentrate on developingan approach to cure a disease for example. The PSEh ss tot andt aine nicali witht areb temh aryb nb isu-a , fore se-q ntedg ech-n nt nalt orei al

environment in which question-based assignments aregiven to users in a simulated laboratory[4]. A subma-rine is launched that immerses the user in the virtualenvironment of the cell populated by sub-cellularcomponents which the user can investigate. With atoolbox, various cellular processes can be investigatedexperimentally. The results of these investigations andexperimentations allow users to solve the assignmentsat their own pace and through their own motivation. Itwas shown that this approach significantly improvesauthentic learning, in particular for large enrollmentgeneral biology classes[5]. At the other end of thespectrum, realistic, fully interactive virtual laborato-ries have been developed to simulate chemical[6],biomedical[7] and recently nanoscience[8] laboratoryexperiments. However, in their present form, these sys-tems were not designed to be integrated PSEs becausethey are targeted to students and researchers for solvingvery specific problems and therefore contain a largeamount of domain specific information. Inexperiencedusers will not have sufficient insight needed for discov-ery and solving problems. Insight, i.e. the capability tomake non-obvious connections between the complexinteractions of the components of these systems, is themain requirement for solving biomedical problems[9]. Such insightful solutions can often be found inan interactive and visual PSE, as demonstrated forexample by the fact that despite the modern numericalcomputing technologies, biophysicists today still use

laradele

ul

is-sen-ry-rld,enea-nslemic-et-a-rialems

as to be scientifically accurate and include accehe state-of-the-art in available data, knowledgeechnology without requiring the user to bring domxpertise and extensive experience with the tech

ntricacies of the PSE that would present the useriresome activation barriers. Novel approacheseing developed, for example a storytelling sysas been presented to fertilize multidiscipliniomedical problem solving[2]. Furthermore, moderiomedical education makes extensive use of vlization of biomedical processes and conceptsxample the publication of the human genomeuence was accompanied by a CD-ROM that preseenome background as well as DNA sequencing tiques in animations[3]. However, these visualizatio

ools are mostly designed to complement traditioeaching techniques and are not very interactive. Mnteraction is provided in the “Virtual Cell”, a virtu

Gedanken experiments for concept development[10].Although there are many virtual reality 3D molecumodels available, biochemists still use hand-mmodels for intuitive reasoning. It is striking that simpintuitive simulation is still one of the most powerfapproaches to creative problem solving.

Since the early days of artificial intelligence,sues of modeling scientific reasoning and its repretation, in particular for those connected with eveday knowledge of the behavior of the physical wohave been studied[11]. At least two aspects have beexplored: multiple representation and qualitative rsoning. Computation with Multiple Representatio(CaMeRa) is a model that simulates human probsolving with multiple representations, including ptures and words[12]. CaMeRa combines a parallel nwork, used to process the low-level pictorial informtion, with rule-based processes in higher-level pictoand verbal reasoning. Furthermore, many AI syst

Y. Cai et al. / Future Generation Computer Systems 21 (2005) 1145–1156 1147

have been developed to simulate the cognition aboutphysical and biological knowledge. What will happenif we spill a glass of milk on the floor? For humans, theanswer is common sense, but understanding this pro-cess is non-trivial for computers. To arrive at an exactsolution, the computer has to solve a set of non-linearpartial differential equations of hydrodynamics that arecomputationally intractable even for simple boundaryconditions[13]. A few studies have focused on thequalitative simulation of physical phenomena. Thus,Gardin uses 2D diagrams to represent physical objectsand their interaction[14] and Forbus uses the fuzzylanguage of “qualitative physics” to model the physi-cal variables[15]. Lower-resolution qualitative modelshave made significant impact in many fields, includ-ing biology. A typical example is the Game of Life, a“Cellular Automaton”[16]. A cellular automaton is anarray of identically programmed automata, or “cells”,which interact with one another. The state of each cellchanges from one generation to the next depending onthe state of its immediate neighbors. By building appro-priate rules, complex behavior can be simulated, rang-ing from the motion of fluids to outbreaks of starfish ona coral reef. Even if the line of cells starts with a ran-dom arrangement of states, the rules force patterns toemerge in life-like behavior. Empirical studies by Wol-fram [17] and others show that even the simple linearautomata behave in ways reminiscent of complex bio-logical systems. In light of this discovery, we intend touse simple biological characters to generate dynamici

eens Mostr mu-l in-s tionssK ona yeesa trix[ thert pled int w”t lems eafl va-t s of

effectiveness and efficiency. Multiphysics simulationand knowledge-based innovation heuristics build thebridge between the innovative idea and reality, whichwill eventually change the landscape of scientific dis-covery. Open source, open system architecture and vir-tual communities bring idea flows from the outside.This is the trend of contemporary scientific discoveryand system design that not only creates just a productbut an inspiration for interaction.

Here we present a computer game as a novel envi-ronment for biological problem solving, where it pro-vides a real-time interactive platform for users. Thegoal of this study is to develop a game-based PSE forusers to explore multi-modal interactions inside a bi-ological system. It includes essential biological sim-ulation models for the immune system and the bloodsystem. It allows users to manipulate and to participatein the interactions among the components of the sys-tem. The biological characters are simulated by soft-ware agents.

2. Prototype

As a test bed for the development of a game-based PSE for biomedical science, we designed ascientific problem that is derived from our ongoingresearch projects. By applying computational lan-guage technologies to the large amounts of wholegenome sequence data publicly available, we havei vela ainstpT findt m-p d tou loodt ody,a thed ad-d s ofa op-i plei eena intot rentc uldb inst

nteractions.Creative problem solving environments have b

tudied mainly in the management science area.ecent studies focus on collaborative creativity, stius culture, and information flow. For example, “bratorming” has been viewed as a panacea in corporaince IDEO has promoted the methodology[18]. Johnao’s “Jamming” theory[19] uses a jazz jam sessis a metaphor to address how to motivate emplot work. Steven Eppinger’s Design Information Ma

20] focuses on representing information flows rahan task flows. According to Constructionism, peoo not simply “get an idea”; they construct it. With

he framework of the recently developed “Idea Floheory to address the dynamics of creative probolving [21], it was shown that the bidirectional idow is the most efficient interaction pattern in innoion. Feedback can enhance the flow both in term

dentified “genome signatures” that may provide nopproaches to the development of vaccines agathogenic microorganisms such as Neisseria[22].he biomedical problem to be explored here is to

reatment for fatal meningitis. In this context, couter scientists without biological background neenderstand the basics of the immune system, the b

ransport system, recognition of bacteria in the bnd possibilities to fight an infection. Furthermore,ynamics of an infection need to be understood, inition to the effectiveness and potential drawbackdministering antibiotics. Within each of the above t

cs, there are a multitude of details to know, for examn the blood transport system, the difference betwctive and passive transport, the points of entry/exit

he tissue, relative dimensions and numbers of diffeell types. Insight into all of these aspects that woe required to begin developing a solution aga


Fig. 1. System hierarchy: the prototype of BioSim involves threelevels of hierarchy (see text).

Neisseria infection would require teaching the funda-mentals of immunology, anatomy, physiology, micro-biology, pharmacology and biochemistry. Thus, tradi-tional teaching tools would offer a high activation bar-rier. The new idea for the biomedical PSE is to developan interactive interface modeled after traditional gameengines that teaches users without background in biol-ogy the understanding of this research problem rangingin hierarchy from atomic to macroscopic scales (Fig. 1).In this hierarchy, the macroscopic level is that of infec-tion of a human body with Neisseria. Fighting the in-fection requires a molecular level understanding of theprocesses involved. The goal is to provide the user withthe necessary insight to creatively generate and testpossible approaches to solving this problem using thePSE.

BioSim version 1.0 is a rapid prototype for this PSE.It is a two-stage game that simulates the journey of redblood cells and white blood cells (macrophages). Thegoal is to introduce the basic concepts of cellular in-teraction and the human immune system. The gamebegins with an animated scene of a blood stream withred and white cells moving passively with the heart-beat. Using a mouse and the arrow keys, the playercan take the role of a biological character, for examplea macrophage, and navigate inside the blood stream.

The game scenario consists of Neisseria cells divid-ing in the tissue at a certain speed. The macrophagescan move actively or passively as specified by the userthrough the bloodstream and need to find and “squeezethrough” the points of exit into the tissue which is onlypossible at the capillaries. Macrophages can then ac-tively move towards Neisseria and “eat” them. Screenshots of these processes are shown inFig. 2.

BioSim 1.0 is implemented on PC. Photorealistic3D models of components of the system were cre-ated with 3D Studio Max and exported to Game Stu-dio 3D (an example is shown inFig. 3). The 3DModeler of GameStudio 3D was used to create thegame scenes, bio-morphed characters and the integra-tion of the world/character dynamics and interactions.C-script, a C-style language, was used to encode thebio-dynamics and bio-sensing behaviors. Game Studiois run under the Windows operation system. It providescapability for either single user or multiple users acrossthe Internet.

2.1. Biological “world model”

In game design, “world models” are similar to thetheatre stage or film scene with which actors and char-acters interact. A world model is often static and largein size. In this project, we developed a comprehensiveworld model that includes the vascular system withartery, veins and capillaries, as well as tissues (Fig. 2Aand B). In this world, the user can fly, walk or runta vel-o lary.T ing”a suei

2

ar-a ria,m rac-t ctersa ec-t plyb tion2 e-f sms,

hrough as one of the biological characters (Fig. 2Cnd D). In the prototype BioSim 1.0, we have deped two scenes: inside and outside of the capilhe transition of the scenes is possible by “squeezcharacter actively from the blood stream to the tis

n the capillary regions.

.2. Biological characters

We have defined the following 3D animated chcters that simulate biological behavior: bacteacrophages, and red blood cells. To define inte

ions among characters and between the charand the world model, we use collision detection (S

ion 2.4.3). For the stand-alone characters, we apio-morphing to assign key frames to them (Sec.4.2). Bio-morphing is accomplished by digitizing d

ormed shapes from microscopic images of organi


Fig. 2. World model and biological characters implemented in the game. (A) Wireframe model of the vascular system world model; (B) 3Dphotorealistic model showing arteries and capillaries; (C) the macrophage (white blood cell) is inside the blood stream with red blood cells; (D)after actively moving out of the blood stream, the macrophage approaches bacteria that infected human body tissue.

building wire frames (Fig. 3) and attaching texture andcolor skins.

2.3. System architecture

The PSE contains three interaction modes: role-play, voyage and networked problem solving.

2.3.1. Role-playThe system allows the user to be a biological charac-

ter in the game. Cognition Science shows that role-playis an important way to stimulate creative ideas. It en-ables the user to have an intimate connection to thecharacter. Also, personalization of a biological charac-ter makes a game more interactive.

2.3.2. VoyageThe user can navigate through the biological system

in the game, either as a character, or using a ‘ship’,supporting different view angles, e.g. traveling throughcapillaries and tissues. Voyage allows exploration atthe user’s chosen leisure, accommodating users withvarious backgrounds.

2.3.3. Distributed problem solvingThe game engine allows users to play the game over

the Internet so that large problems can be solved collab-oratively or antagonistically, e.g. some users can playmacrophages and others can play bacteria. The dis-tributed problem solving enables diverse game strate-gies and more excitement of the game. The user canalso choose between two aims, rather than playing therole of a single biological component only. The usercan assume the roles of multiple biological characters,thus studying their individual influence on a particularaim. These aims are to induce an infection with Neisse-ria and ensure its successful propagation in the humanbody or to fight the Neisseria infection.

2.4. Biological character library

A biological character is defined as an entity thatincludes functions, forms, behaviors and interfaces.For example, some of the functions of a macrophageare to locate and destroy bacteria (see below), whilethe function of a bacterium is to divide and spread.A given character has a sequence of forms in cor-respondence to its dynamic behaviors. An interface


Fig. 3. 3D modeling of biological characters. A wireframe is con-structed using sufficient numbers of polygons to allow realistic bio-morphing of characters, here a bacterium.

to interact with other entities or the environment isdesigned.

Example macrophage:

• Functions: locate bacteria| destroy bacteria.• Behaviors: walk| eat| shrink| kill | die.• Forms (key-frame sequences): moving| search|

squeeze| killing | death.• Interfaces: collision| chemotaxis (sensing)| tactile

(sensing).

The principles describing behaviors, forms and in-terfaces that are common to all characters are describedbelow.

2.4.1. BehaviorInteraction is the key to computer games, and we

believe also to an efficient biomedical PSE. We there-fore allow the user to control the behavior of biologicalcharacters by realistic and scientifically accurate bio-interactions. The transitions of each character are rep-resented by a state machine (Fig. 4A). For example amacrophage’s states include the transitions to deform,shrink, eat, walk and die. Each behavior is defined by aset of forms (see Section2.4.2). We simulate biologicalprocesses and character behavior (“bio-dynamics”) re-alistically. For each character, we define its interactivemodes, such as motion, reproduction and death. Tak-ing bacteria for example, we use the following rules(Fig. 4B):Autonomous motion. Given a fixed duration, each

bacterium moves a distancex at angley, wherex andyare random values. The distance should not exceed themaximal distance.Reproduction. Given a predefined duration, each

bacterium reproduces a copy of its own which is placedbeside its original position. TheLogistic growth model[23] adequately describes the reproduction process ofsimple organisms over limited time periods by Eq.(1),whereM is the carrying capacity of the population Ba

B

D llse

Fig. 4. Examples for character dynamics. (A) State machine for a ma Automata;(C) example for the time-evolution of bacterial growth and spatial dist

t timenwith growth rater:

n+1 = Bn + rBn

(1 − Bn

M

). (1)

eath. If a bacterium’s life cycle is over or if other ceat it, it is removed from the scene.

crophage; (B) dynamics of an organism modeled by Cellularribution.


Fig. 5. Realistic and automatic generation of changes in the cellular forms. Cellular images from time-resolved microscopy studies[27] areanalyzed using active contour modeling[24] and 3D models are generated.

2.4.2. Form: shape tracking and bio-morphingTo give users precise control and accurate feel of a

character’s movements, vivid animation sequences aregenerated for each character. Our approach is to usecomputer vision-based shape tracking methods to au-tomatically detect the changes in contours of movingcells from microscopy studies. We use active contourtracking models[24], also called snakes, since theyare more robust in noisy background typical for bio-logical images than traditional edge detection meth-ods. Active Contour Modeling provides a general al-gorithm for matching a deformable model to an imageby means of energy minimization. The energy func-tion is a weighted combination of internal and externalforces. The active contour is defined parametrically asv(s) = [x(s), y(s)], wherex(s), y(s) arex, y coordinatesands∈ [0,1]. The energy function is

Esnake=∫ 1

0Esnake(v(s)) ds

=∫ 1

0{[Eint(v(s))] + [Eimage(v(s))]

+ [Econ(v(s))]} ds (2)

whereEint represents the internal energy of the splinedue to bending,Eimage the image forces, andEcon theexternal constraint forces. Usually,v(s) is approxi-mated as a spline to ensure desirable properties of con-tinuity.Fig. 5shows an example, the moving patterns ofa usingt es,

we build a 3D model for a character using polygones(as shown inFig. 3for a bacterium).

2.4.3. InterfaceWe have implemented three types of interfaces: col-

lision detection, chemotaxis sensing and tactile con-trol. Collision detection describes how two objects be-have when they collide. Chemotaxis sensing describeshow an organism senses environmental stimuli. For

Fig. 6. Collision detection. (A) Boxes surround objects when coarser , buto

macrophage. The macrophage area is croppedhe active contour model. Based on the outline fram

esolution is sufficient. Polygons are used for higher resolutionbjects may not be concave (B), but must be convex (C).


example, white blood cells can “smell” bacteria andmove toward them by chemotaxis. Tactile control addsphysical properties to the object. For example, an ob-ject can be very ‘sticky’ to the mouse movement, or theobject can be very heavy so that the user has to click theleft button a few times before its movement happens.Collision detectionis implemented based on the res-

olution required as a balance between speed and accu-racy (Fig. 6): coarse collision detection using boxesis implemented when red blood cells passively movethrough the blood stream. Tight collision detection us-ing polygons is used when high resolution is necessary,for example during the interaction between bacteria andmacrophages. The 3D game studio currently only sup-ports convex map entities. Concave objects (objectsthat have indents) must be converted to convex map

entities. This is challenging in constructing the worldmodel.

To simulatesensing capabilities, we define a circu-lar envelope around the character as a sensing region.When the target is inside the region, the character willmove towards the target and engage in interactions withit. To enable more complex means of interaction withthe biological world, each user is equipped with a ‘ship’that allows for effective captivation by the environ-ment (Fig. 7, upper left). The ‘ship’ provides a meansof transportation (Voyage) and action (Role Playing).Each activity is determined by availability of “energypoints”, which have to be carefully balanced to min-imize consumption and maximize effectiveness. Theuser knows the status of energy points via a controlpanel, which also provides for the various possibilities

F anspor e ship hass ria with can divideu the his n mark them( ing, th . In futurei ultiple gle game.Ssd

ig. 7. Captivation of the user by the PSE. A ship provides trensing and action capabilities (upper right). Marking of bactendisturbed (1, 2). When the ship approaches the bacteria (3),in the game this is indicated by a change in color). After markmplementations of the game, internet connectivity will allow m
ince ships are not-self, the human immune system would identify tystem tools, i.e. antibodies (lower right). This is represented by theirect macrophages to “eat” this ship.
tation through the body (upper left). A control panel inside thhistamines (lower left). Bacteria that have not been marked

tamine sensor identifies the bacterial infection and the user cae user can attract macrophages that will “eat” the bacteria (4)users to participate in collaboration or as antagonists in a sin
hem as such. Therefore, one ship can mark another ship using immunehalo of one of the ships. Once labeled with antibodies, the user can again


of action (Fig. 7, upper right). For example, in a stateof high energy, the user can afford to travel activelywith the ship to a point of infection. However, in a stateof low energy, the user would choose to travel pas-sively with the blood stream. This will allow the userto further develop decision-making skills in a biolog-ical PSE. In the future, there will also be additionalbio-sensing capabilities available through the controlpanel, for example a histamine sensor (Fig. 7, lowerleft) and mechanisms of the immune system to distin-guish self from non-self (Fig. 7, lower right). This willallow introduction of molecular level information, forexample the user will need to use molecular docking ofthe immune system’s antibody structures to those of thebacterial surface structures. This will train users to viewprotein structures and understand the mechanisms ofcomplementarities of two structures. The player seek-ing to evade the immune system would need to developstrategies to evade antibody marking, e.g. through sur-face mutation. Thinking about possible strategies fromeach point of view will allow the user to gain deep in-sight into the factors controlling the health of the organ-ism, from the molecular to the macroscopic level, ulti-mately aiding in the development of novel solutions forbiomedical problems such as the Neisseria infection.

3. Problem solving with BioSim

3

s oft ssues

in biomedical research on users with no background.The ideal group at this stage of implementation of thegame is young children, for two reasons. One, childrenare unbiased and without background. Second, childrenlearn optimally when the material to be learned is pre-sented to them in an accurate way to avoid the build-upof incorrect models by implicit learning[25]. Implicitlearning of correct biomedical concepts by childrentherefore requires the same fundamental issue of scien-tific accuracy as other users will require once the gamereaches the stage of providing a PSE for users with anybackground. We tested BioSim 1.0 on 14 children atKinderCare, Cranberry, PA, on 9 August 2002 and 25February 2003. We let 4- and 5-year-old children playwith the game on a laptop and focused our attention onstrategic aspects and active questioning.

Two strategies were quantified, the speed ofmacrophage movement towards the bacteria (Fig. 8,left) and the use of antibiotics in aiding the killing ofthe bacteria (Fig. 8, right). All children learnt quicklyto shift from fast pace chasing to slow pace chas-ing so that their capture rate was improved. We thentested a more challenging concept, that of usage ofantibiotics to aid the killing of the bacteria. We in-cluded the ability of bacteria to develop resistance inour growth model. Thus, the children had to discoverthat antibiotics at some stage in the game no longerinhibit bacterial growth. This was only observed by asingle 5-year-old, all other children kept on adminis-tering antibiotics despite energy consumption and lacko vea novelw ills

F d in the iotics (right).T pace be limited byg inistra

.1. Supporting strategic thinking in users

We conducted experiments in the effectiveneshe game to raise an awareness of the important i

ig. 8. Strategies against bacterial infection that can be explorehe fast pace strategy often leads to missing targets. The slowradual development of bacterial resistance. At that point, adm

f effect (Fig. 8, right). These types of quantitatissessment of strategic behavior of users openays to analyze learning of problem solving sk

game, speed of macrophage movement (left) and use of antibstrategy gains steady capture rate. The use of antibiotics cantion of drug does not inhibit bacterial growth.


Table 1Comparison of the reactions of two groups of children to the game

Observation 4-Year-old 5-Year-old

Asked relevant questions 0 4Controlled the game successfully 2 5Described bacterial growth behavior 1 5Described macrophage behavior 2 6

The children in the first group were 4 years old, those in the otherwere 5 years old. Each group consisted of 7 children, and the totalnumber of children tested was 14.

that would not be possible with conventional teachingmethods.

3.2. Supporting creative thinking in users

To assess learning in children, we asked questionssuch as “How does the macrophage get out of thecapillary?” or “How do you kill bacteria?” The chil-

dren used intuitive metaphors, for example the anal-ogy of “vacuum” and “crash into” to describe how themacrophage attacks bacteria. This shows that the play-ers are very sensitive to the intimate design details ofthe game, which opens a window for game develop-ers toencodevery subtle knowledge about complexbiological interactions. Finally, we tested the game-induced stimulation of questioning and creative think-ing in the children. The results are summarized inTable 1. The 5-year-old children asked several mean-ingful questions, for example: “Are bacteria germs?”,“Where do the white cells go?”, “What’s a red cell?”,“Where do the bacteria live?”, “Is the macrophagegood or bad?”. Overall, 4-year-old children asked fewerquestions, and most of their questions were not rel-evant, for example, “Do you have other games?”, “Idon’t want my head eaten off”. These observations sug-gest that there may be a turning point between ages 4and 5 where a PSE can become effective. Studying

assistin
Fig. 9. Contradiction matrix for g problem solving using TRIZ[26].


these reactions to BioSim in more detail with a largernumber of children of varying ages is a major futuregoal.

3.3. Supporting innovative problem solving

Encouraged by the results on stimulating activethinking in the users, we are now in the process of inte-grating BioSim with the TRIZ model (the acronym for“inventive problem solving” in Russian) to prepare forthe next step, the use of BioSim for innovative problemsolving (Fig. 9). TRIZ was originally developed by Alt-shuller[26] based on the hypothesis that innovation in aparticular field can benefit from the general innovationexperience from other areas through problem-solutionanalogy, and there are ways to extract these innovationprinciples. Over 2 million patents were examined, clas-sified by level of inventiveness, and analyzed to lookfor principles of physical problem solving. We haveimplemented Altshuller’s Contradiction Matrix in anexternal utility in Java. The matrix catalogues the pa-rameters on theX- andY-axes:

• X[1–39] = [weight of a mobile object| energy spentby a moving object. . .].

• Y[1–39] = [speed| force| temperature| shape| pres-sure. . .].

The matrix’s cells (1521 except diagonal) containprinciples which should be considered for simultane-ous optimization of parameters. For example, to im-p ver-s vingo ial),1 ro-p esep arep ed-i

4

tra-d s inc tiona di-c inesa rtifi-

cial intelligence technologies and creative instructiontechnologies. In this PSE, cross-disciplinary educationwill be on-demand, entertaining and interactive. Thiswill allow focus on discovery and creativity rather thanone-way tutoring. Towards this long-term goal, here,we have presented a game-based PSE, where users canexplore complex biological interactions with naviga-tion, role-play, and networked collaboration. The studyinvestigates the system architecture of the biologicalgame, bio-morphing characters, and bio-interactionswith bio-sensing and bio-dynamics. The game is basedon realistic biological models, such as logistic growthmodels of simple organism reproduction and immigra-tion models of cell movements. The prototype has beenimplemented on PC and tested in a preschool environ-ment where users have little knowledge in biology. Theexperiment shows that the game greatly inspires usersboth in concept learning and entertainment. It supportsstrategic and creative thinking, as well as innovativeproblem solving.

Acknowledgements

This work was supported by the Alexander vonHumboldt-Foundation and Zukunftsinvestitionspro-gramm der Bundesregierung, Germany.

References

ch-ogy,nceingca-

L.in-Pro-

anngsand

.R.nif-

12thning,ack-

rove the intersection volume (the 7th parameter)us energy spent (the 19th parameter) of a mobject refer to the principal number 12 (Equipotent8 (Vibration), and 31 (Porous Materials) as appriate concepts to build up solutions. Currently, thrinciples are limited to the physics domain, and welanning to expand the knowledge base to the biom

cal domain.

. Summary

Future biomedical problem solving is beyonditional means because of the existing challengeross-disciplinary communication and interpretand utilization of vast quantities of available biomeal data. We want to build a virtual PSE that combdvanced computer graphics, computer vision, a

[1] M.C. Roco, W.S. Bainbridge, Overview: converging tenologies for improving human performance: NanotechnolBiotechnology, Information Technology, and Cognitive Scie(NBIC), in: M.C. Roco, W.S. Bainbridge (Eds.), ConvergTechnologies for Improving Human Performance, Kluwer Ademic Publishers, Arlington, 2002, pp. 1–23.

[2] A. Kuchinsky, K. Graham, D. Moh, A. Adler, K. Babaria, M.Creech, Biological storytelling: a software tool for biologicalformation organization based upon narrative structure, in:ceedings of the AVI, 2002.

[3] http://www.sciencemag.org.[4] A.R. White, P.E. McClean, B.M. Slator, The virtual cell:

interactive, virtual environment for cell biology, in: Proceediof the Wold Conference on Educational Media, HypermediaTelecommunications, Seattle, WA, 1999.

[5] P.E. McClean, B. Saini-Eidukat, D. Schwert, B.M. Slator, AWhite, Virtual worlds in large enrollment science classes sigicantly improve authentic learning, in: Proceedings of theInternational Conference on College Teaching and LearCenter for the Advancement of Teaching and Learning, Jsonville, FL, 2001.


[6] http://ir.chem.cmu.edu/irproject/applets/virtuallab/.[7] http://www.biointeractive.org/.[8] M. Guggisberg, P. Fornaro, T. Gyalog, H. Burkhart, An interdis-

ciplinary virtual laboratory on nanoscience, Fut. Gen. Comput.Syst. 19 (2003) 133–141.

[9] Y. Cai, Pictorial thinking, J. Cogn. Sci. (1986).[10] T. Hayden, The inner Einstein, in: US News, 2002.[11] J. McCarthy, P.J. Hayes, Some philosophical problems from

the standpoint of artificial intelligence, in: D. Michie (Ed.),Machine Intelligence, Edinburgh University Press, Edinburgh,1969.

[12] H. Simon, Models of Thought, vol. II, Yale Press, Yale, 1989.[13] R. Feynman, The Feynman Lectures on Physics, Addison-

Wesley, Reading, Massachusetts, 1963.[14] F. Gardin, B. Meltzer, Analogical representations of naive

physics, Artif. Intell. 38 (1989) 139–159.[15] K.D. Forbus, Qualitative process theory, Artif. Intell. 24 (1994)

85–168.[16] M. Gardner, Mathematical games: the fantastic combinations of

John Conway’s new solitaire game “Life”, Sci. Am. 223 (1970)120–123.

[17] S. Wolfram, A New Kind of Science, Wolfram Publishing, 2002.[18] M. Schrage, Serious Play: How the World’s Best Companies

Simulate to Innovate, Harvard Business School Press, Boston,2000.

[19] J. Kao, Jamming, HarperBusiness, 1997.[20] S.D. Eppinger, Innovation at the speed of information, Harvard

Business Rev. 79 (1) (2001) 149–158.[21] Y. Cai, Idea flow, in: Proceedings of the International Confer-

ence of Design and Management, Boston, 2001.[22] M. Ganapathiraju, D. Weisser, R. Rosenfeld, J. Carbonell, R.

Reddy, J. Klein-Seetharaman, Comparative N-Gram Analysisof Whole-genome Protein Sequences, Human Language Tech-nologies, San Diego, 2002.

[23] A.C. Guyton, J.E. Hall, Textbook of Medical Physiology, Saun-

[ tour

[ en-iss J.

[ ion

[ ar-

ser-ci-te

Re-ry,i-edb,ing

Ingo Snel was born and lives in Frank-furt, Germany. He was trained as a plumberbefore he discovered 3D computer designwhich is now his full-time occupation. Inthe BioSim project he is responsible for thedesign, texturizing and animation of char-acters and scenes.

Betty Cheng is a Master’s student in theLanguage Technologies Institute at theSchool of Computer Science at CarnegieMellon University. She received herbachelor’s degree in Computer Science andMathematics from Simon Fraser University,Canada. Her research interests lie in theapplication of language technologies to pro-teomics. She plans to pursue a PhD degreein bioinformatics, specializing in protein–protein interactions and new drug design.

B. Suman Bharathi is a Master’s studentin Medical Microbiology at Baroda Univer-sity, India. She is working on identificationof microbial genome signatures as part ofher Master’s thesis research project at Re-search Center Julich (Germany).

Clementine Klein was born in Opladen,Germany. She studied Art and Design atthe college of Arts, Cologne and worked

er.lit-ce

tcol-olp-

m-er-So-dt-heter

J as ofb utureo ss-d

ders, Philadelphia, 1997.24] M. Kass, A.P. Witkin, D. Terzopoulos, Snakes: active con

models, Int. J. Comput. Vision 1 (4) (1988) 321–331.25] P. Perrig, W.J. Perrig, Implicit and explicit memory in m

tally retarded, learning disabled, and normal children, SwPsychol. 54 (2) (1995) 77–86.

26] G. Altshuller, The Innovation Algorithm, Technical InnovatCenter, 2000.

27] J.A. Sullivan, Chemotaxis of Human Neutrophils, Chlottesville, VA, USA.

Yang Cai is a faculty member (SystemScientist) at the Human–Computer Intaction Institute, School of Computer Sence, adjunct faculty member of Institufor Complex Engineered Systems, andsearch Fellow of Studio for Creative InquiCollege of Fine Arts, Carnegie Mellon Unversity. He is also a founder of AdvancImaging and Modeling Lab/Biovision Lawhich explores the Perceptive Computproducts.

from 1972 to 1975 as a graphic-designShe then studied biology and Germanerature at the University of Cologne. Sin1982, she works as an instructor.

Judith Klein-Seetharaman is AssistanProfessor at the Department of Pharmaogy at the University of Pittsburgh Schoof Medicine (USA) and holds secondary apointments at the Research Center Julich(Germany) and at the School of Coputer Science at Carnegie Mellon Univsity (USA). She has been awarded thefya Kovalevskaya Prize by the HumbolFoundation, with host institutions GoetUniversitat Frankfurt and Research Cen

ulich (Germany). Her primary research interests are in the areiochemistry, biophysics and bioinformatics. She envisions a ff multiple virtual problem solving environments supporting croisciplinary education, collaboration and research.

Documents

BioSim—a biomedical character-based problem solving ... WEBSITE/images/publications/BioSim … · ronment for biological problem solving, where it pro-vides a real-time interactive